search

taotaowang97479 / MFNet-SpeechEnhancement

This is the unofficial implementation of MFNet, from paper''a Mask Free Neural Network for Monaural Speech Enhancement''
10 stars 0 forks source link

question about input-output of MFNet #3

Open JangyeonKim opened 6 months ago

JangyeonKim commented 6 months ago

Thank you for your code. I have a question. The input to the model consists of compressed spectra, but when calculating the loss, is it compared with the original STDCT spectra? In other words, is the input compressed but the output expected to be the original spectra?

JangyeonKim commented 6 months ago

I'm confused about whether to use the original input spectrum for calculating the loss and during the mapping reverse noise process, although the input is compressed.

taotaowang97479 commented 6 months ago

I use the compressed spectrum during the training process, including the calculation of losses, that is, the input and output are compressed spectra. Also, I didn't use reverse noise.